Baseten is assembling a round of about $2B that would price the San Francisco inference specialist between $26B and $30B, people familiar with the talks told Bloomberg. Existing investor Sands Capital is expected to lead.
The structure is unusual: the financing is split into two tranches, the first setting a $26B mark and the second stretching to $30B, suggesting the round depends on milestones rather than a single negotiated number.
Either way, it is a steep step up. Baseten closed a Series F of $1.5B in June that valued the company at $13B. A jump to $30B inside five months would rank among the fastest re-ratings of the current cycle.
Baseten, founded in 2019, sells software for running models in production once training finishes. Low latency, steady availability and predictable cost per token are the selling points, and Nvidia has used the company in case studies about cloud-scale inference.
The momentum reflects where AI spend is travelling. Training budgets cluster around a handful of labs, while inference touches every app that ships a model to users. Investors have been writing larger cheques for the serving layer accordingly.
Powerlattice, Firehawk and FuriosaAI have all banked significant sums in recent days, each attacking a different bottleneck in the same buildout. Multiples like $30B are the clearest signal yet that capital is chasing capacity and delivery rather than research ambition.
What remains unanswered is margin. Inference is a commodity market with thin differentiation on price, and Baseten will need to hold customers as cloud providers bundle competing services into their own catalogues.
